| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
| |
|
| |
|
|
|
|
| |
It overcomplicates things.
|
| |
|
|
|
|
|
|
|
| |
We run into a race condition when testing in parallel when all
processors install test gems to their test folder (due to shared gem
cache). So I figure we can instead run this setup sequentially
beforehand.
|
| |
|
|
|
|
|
| |
Regular builds shouldn't need any extra configuration deviating from a
standard ubuntu setup.
|
|
|
|
|
|
| |
That version of groff is already installed.
As a result, the previous `apt-get update` is not needed either.
|
|
|
|
|
| |
Hopefully `rvm` won't install this gem by default, but for now
I'm adding the following two methods for compatibility with it.
|
|
|
|
|
|
|
|
|
|
|
| |
Sometimes `rake` will load the default version of `bundler` and then the
`Rakefile` will change the LOAD_PATH and require the same code again
using the version of `bundler` in `lib`.
The only `bundler` code that gets used by the `Rakefile` is
`bundler/gem_tasks`. This is used, for example, for releasing. I think
it's fine to use an installed version of `bundler` for this in any
case.,
|
|
|
|
|
| |
We vendorize it as a dependency of `net-http-persistent`, so usages
inside `net-http-persistent` get automatically replaced by `automatiek`.
|
| |
|
| |
|
| |
|
|
|
|
| |
We never use system rubygems anymore. Always the local clone.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Instead, use a better rubygems version manager that makes sure that
`ENV["RGV"]` usage is always applied, and its usage is centralized at a
single place.
This simplifies the rake task organization and makes it much simpler to
run locally exactly what's run in CI. For example, since it doesn't rely
on the `RGV` being deleted for subsequent subprocesses, it also running
`bin/rake spec:travis` task locally which previously would lead to
```
rake aborted!
RubyGems version is required on Travis!
```
It also reduces duplication making it so that we only need to change the
`.travis.yml` file when updating tested rubies, and not the `Rakefile`
as well.
|
| |
|
|\
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
7342: Start using automatiek 0.3.0 r=deivid-rodriguez a=deivid-rodriguez
### What was the end-user problem that led to this PR?
The problem was that I added a feature to automatiek to vendor subdependencies, but we were not using it because it had not yet been released.
### What is your fix for the problem, implemented in this PR?
My fix is to start using the feature now that it has been released.
### Why did you choose this fix out of the possible options?
I chose this fix because it allows us to manage the `net-http-persistent` dependency in an easier way.
Co-authored-by: David Rodríguez <deivid.rodriguez@riseup.net>
|
| | |
|
|/
|
|
|
| |
If you don't have `ronn` installed, the `man:check` task will fail but
it won't let you know the exact reason.
|
|
|
|
| |
Since it still has issues.
|
| |
|
|\
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
7274: Fix more leaks to default copy of bundler r=hsbt a=deivid-rodriguez
### What was the end-user problem that led to this PR?
The problem was that in some places, it was still possible to end up requiring files in a different copy of bundler (the default copy). I noticed this when I removed a rubygems monkeypatch from the test suite that was preventing the default copy of bundler from being activated when requiring files.
This thing:
https://github.com/bundler/bundler/blob/e1c518363641208429f397170354054b3d28effd/spec/support/hax.rb#L15-L20
### What was your diagnosis of the problem?
My diagnosis was that I should use relative requires wherever they were missing.
### What is your fix for the problem, implemented in this PR?
My fix is to remove the rubygems hack, migrate the rest of the internal requires to be relative, and also introduce some hacks on our specs to make sure we never load the incorrect copy of bundler.
I think this PR should fix the issues in https://github.com/rubygems/rubygems/pull/2863.
Co-authored-by: David Rodríguez <deivid.rodriguez@riseup.net>
|
| | |
|
|/ |
|
|
|
|
|
|
|
|
| |
Different versions of `groff` seem to generate slight differences in
whitespace. It is too much burden for contributors in my opinion to make
them install the exact groff version CI is running, and whitespace
amount in man pages doesn't seem to change the rendered output anyways,
so it should be good I think.
|
|
|
|
|
| |
To make it more predictable and easier to find out about discrepancies
in the output.
|
|
|
|
| |
To try fix `man:check` failing sometimes.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This has the benefit that:
* Allows the installation of bundler as a default gem from rubygems to
include man pages.
* Removes the need to build man pages during our tests.
* Makes working with the manifest easier, because we only have source
controlled files, and not a mix of source control and generated files.
To make sure they never fall out of sync, we replace the previous
`man:build` CI task with a `man:check` task that makes sure the
generated man pages are up to date.
|
|
|
|
| |
The build task does not generate any files at `lib/bundler/generated`.
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
* Adds an extra artifice task to vendorize new `connection_pool`
dependency.
* Cherry-pick's needed Windows fix not yet merged into master branch of
`net-http-persistent`.
* Update bundler usages to be compatible with the new version, and fix
unit specs.
|
|
|
|
| |
And refactor development setup.
|
| |
|
| |
|
|\
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
7223: RUBY_ENGINE should always be defined after 1.8.7 r=deivid-rodriguez a=deivid-rodriguez
### What was the end-user problem that led to this PR?
The problem was that we have code checking that `RUBY_ENGINE` everytime it's used and I think it's unnecessary.
### What was your diagnosis of the problem?
My diagnosis was that every ruby implementation newer than 1.8 should define this.
### What is your fix for the problem, implemented in this PR?
My fix is to remove the unnecessary code.
Co-authored-by: David Rodríguez <deivid.rodriguez@riseup.net>
|
| | |
|
|/ |
|
|
|
|
|
| |
And move its description on top of the correct task, `man:build`, so
that it appears under `bin/rake -T`.
|
|
|
|
|
| |
Instead of creating a separate rake task. This should allow the tests to
properly clean after themselves also when not run through rake.
|
|
|
|
|
|
|
|
| |
The current numbers are misleading and coverage tracking doesn't work in
most of the places since the current test suite is mostly based on
spawning subprocesses, and coverage doesn't get properly tracked there.
Let's revisit this in the future.
|
|
|
|
| |
So that `rake install:local` at least works.
|
|
|
|
|
| |
The ronn gem is needed for building man pages, that's already guarded.
The `release` task is irrelevant.
|
|\
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
7070: Improve cross repo integration r=deivid-rodriguez a=deivid-rodriguez
### What was the end-user problem that led to this PR?
The problem was that we had on `rubygems`, we're not testing bundler against the rubygems checkout of each PR, but against master. That is confusing because developers assume that the PRs tests will run against the PRs code, but this is currently not true.
### What was your diagnosis of the problem?
My diagnosis was that on rubygems we are [specifying `RGV=master`](https://github.com/rubygems/rubygems/blob/fa6e547330e97b417ed11f262d68d03c57abeeda/.travis.yml#L12) in our TravisCI matrix. And what `bundler` does with that is to clone rubygems master under `tmp/rubygems` and use that. Thus your rubygems PR is not really tested.
### What is your fix for the problem, implemented in this PR?
My fix is to rework the setup to accept an `RGV` environment variable when it contains. In that case, no git operations will be performed, and that path will be used directly and assumed to contain a rubygems checkout.
I also refactored a few things about the setup while at it.
### Why did you choose this fix out of the possible options?
There were existing tasks to test bundler against a rubygems checkout, but they were unused and they expected a `RG` environment variable instead. I decided to reuse these tasks, but make them pick up the `RGV` as long as it contains a valid path. I chose this fix because it works and it doesn't make the existing setup too different.
Co-authored-by: David Rodríguez <deivid.rodriguez@riseup.net>
|
| | |
|
| | |
|
| | |
|